If AI has a leader, what skills should it have? It's rare that AI is racing in two extreme directions at the same time. On the one hand, generative AI seems to be omnipotent and constantly integrates and unifies; Another aspect is that data, algorithms, and computing resources seem to be increasingly isolated and difficult to trust. How to put it, so orderly yet chaotic and torn apart, the last time I saw it was the Five Dynasties and Ten Kingdoms. Hey, pull the sheep? Let's sneak in! ⬇️ There are also many solutions on the market to optimize inference results. Compared to Bittensor's traditional model fusion method of "scoring voting+average aggregation", @ AlloraNetwork uses a more clever "model peer evaluation" mechanism. Simply put, it does not rely on a group of judges to rate a model and then calculate an average, but rather allows each model to evaluate the error performance of other models and dynamically allocate weights based on these evaluations. In this way, the entire network is like an intelligent team that can learn and improve on its own, and models that perform better will gain greater say. For example, suppose a group of people are guessing tomorrow's weather, the Bittensor method may be to take the average of the numbers reported by everyone. Allora is like a mind reading expert who not only looks at everyone's guesses, but also analyzes who is more reliable and prone to errors, and dynamically adjusts everyone's weights to make the final prediction closer to the real situation. According to the simulation test of the Allora white paper, this method has a very significant effect: compared with traditional fusion methods, Allora can reduce the overall error to one tenth or even lower than the original. / What exactly does Allora want? Allora advocates creating a decentralized AI network that can evolve on its own, utilizing machine learning models built by communities for highly accurate and context aware predictions. When it comes to human language, Allora can do big things and become the leader. They use blockchain to unify and schedule N models for collaborative reasoning, building a smarter, more transparent, and secure intelligent system, which they call the AI abstraction layer. Expanding upon it is a new knowledge point, MCN - Model Coordination Network， Let me translate it as' multi model coordination network 'for now. Allora is currently the only network on the market that can dynamically integrate multiple models into a context aware inference system. The amount of information is too dense, let's start with a brief introduction and understand the process. How did Allora do it? I have read the white paper several times and also went to the blog to read various cases. It was overwhelming and there were too many mathematical formulas, but in fact, it can still be abstractly summarized. The Allora processing process that IQ 50 can understand should be like this: For example, if I want to obtain the housing price forecast results, I can create a theme on Allora to predict the housing prices in Shanghai Q3, with a unit of RMB and an error of 1%. At this point, various roles and participants in the network can enter: real estate agents provide a model, economists provide a model, government officials provide a model, and evaluators pledge tokens for accuracy verification; In the end, those who need this data can pay tokens to purchase the predicted results. It doesn't seem difficult to understand either? I also summarized the process of IQ 100 as follows: Theme creation Participants join Inference generation Reasoning synthesis Verification and rewards Self evolution Delivery of Results Comparing with Bittensor again, you will find that Allora actually focuses on generating collective intelligence beyond individual participants, generating high-precision inference through context aware "Inference Synthesis". It can cover a wide range of complex AI tasks such as regression and prediction. // It is interesting that Allora recently published an article detailing 31 cryptocurrency scenarios that they can apply, such as DeFi Agents, predictive markets, perpetual contracts, asset management, dynamic insurance pricing, automated market makers, social sentiment analysis, supply chain optimization, residual systems, medical forecasting, game consumption optimization, and more. Does this example make it clear? By examining our partners, we can also see what Allora can do. On the homepage of the official website, there are AWS and Alibaba Cloud, and clicking on them will reveal many familiar names:@ OpenledgerHQ 、Spheron、Soneium、ElizaOS、Virtuals Protocol、Coinbase AgentKit …… As of today, Allora has achieved 692 million intelligent inferences, 288000 participation models, 55 task themes, and has established cooperative relationships with over 50 projects, of which 27 projects have achieved real-time integration and generated revenue directly from the mainnet launch. Digging deeper, Allora was founded in 2019 and has received $35 million in investment from top VC firms such as Polychain and Framework Ventures, keeping a low profile. This is another practical person who doesn't need PPT to start a business. They already have real customers and can generate real profits by launching on the main website. This is the direction I have been looking for. Allora， It can be included in the key focus list, looking for opportunities from the ecology and upcoming secondary markets. This is a soft core science popularization article, through which you can have a partial understanding of the following knowledge: What is MCN How Allora coordinates multiple model networks Bittensor is just a younger brother (not Author: Anymose | A Soft Core Science Popularization Writer<Full Text End> *This article is for educational purposes only and does not constitute any investment advice. Always remember DYOR!